Documentation
LLM Hangar deploys open large language models onto GPU infrastructure in your own cloud account and gives you a private, key-authenticated, OpenAI-compatible endpoint. You can manage the resources in your provider console. Your provider bills you directly under your own agreement.
Quickstart
- Create an account and start the 7-day free trial. No credit card is needed.
- Connect a cloud account: AWS, Nebius, RunPod, or Verda.
- Deploy a model from the catalog. Pick a GPU shape and region, review the cost estimate, confirm. For typed decisions instead of text, see decision models.
- When the deployment is ready, open its Connect tab for the endpoint URL and API key.
- Point any OpenAI-compatible client at it.
Guides
AWS via a CloudFormation cross-account role, Nebius via a project-scoped service account, RunPod via an API key, Verda via client credentials. What each needs and what we verify.
The deployment flow from catalog to running endpoint: shapes, regions, cost estimates, and what happens during a boot.
Working snippets for Python, curl, JavaScript, LangChain, the Vercel AI SDK, editors, and automation tools.
Deploy Laya for typed choice, score and yes or no answers with probabilities, and call it with the TypeSafe clients, curl or any HTTP tool.
How budget estimates, deletion timers, schedules and teardown checks help you manage costs.
States including stop_failed, the deploy-time staging choice and its cost, wake estimates, what a stopped endpoint answers, and the API.
Configure weekly windows during deployment or later, with overnight runs, time zones, savings previews and pre-warm, holds and wakes, webhook deliveries and signature verification.
Eviction per provider, automatic recovery from staged weights, the fallback and wait policies, and reacquiring spot.
Named keys with limits, JWT validation against your identity provider, and audit capture to a bucket you own with chains you can verify.
Connect your Verda project and manage instances, staged weights and startup scripts.
Which EC2 instance fits which model, prices by region with their sources, the quota and capacity blockers, and the four steps.
Where prompts go, how access to your cloud account is scoped, credential storage, the audit log, verified teardown.